Smart-Thinking

Smart-Thinking is a local, deterministic MCP server that builds and analyzes a reasoning graph entirely on your machine, without relying on external AI APIs. It uses a local TF-IDF similarity engine, heuristic-based scoring, verification tracking, and a persistent memory/ledger to produce transparent MCP responses and a timeline of reasoning steps. To interact with it, install the MCP package globally or run it via npx, then feed thoughts through the reasoning pipeline. The built-in CLI demonstrates a complete reasoning workflow, from session bootstrap to persistence and visualization of the decision path. The server supports different clients via connector mode and ships with a demo that shows how nodes are added, heuristics evaluated, and verification feedback recorded in sequence.

Key tools and capabilities include: a graph-first reasoning model with rich relationships between thoughts; a local memory store and memory manager for persistent sessions; a heuristic evaluator and metrics calculator that produce traceable scores; and a verification service that annotates nodes with statuses and calculation traces. For client compatibility, you can run the server in connector mode when a client only needs search and fetch tooling, or run the full toolset locally for interactive reasoning and debugging.

Prerequisites:

• Node.js (recommended latest Active LTS) and npm installed on your platform
• Git (for source installation, optional)

Option A: Global installation (recommended)

• Install the MCP server globally so you can run it from anywhere:

npm install -g smart-thinking-mcp

• Run the server with npx (no local install required per run):

npx -y smart-thinking-mcp

Option B: From source

• Clone the repository and install/build locally:

git clone https://github.com/Leghis/Smart-Thinking.git
cd Smart-Thinking
npm install
npm run build
npm link

• Then start the server from source using the locally linked package or npm scripts as documented in the repository.

Note: If you need platform-specific setup details, refer to GUIDE_INSTALLATION.md in the project.

Tips and common considerations:

• Smart-Thinking runs fully locally by default (Zero API Key mode). External verification/search connectors are disabled unless you explicitly enable them via SMART_THINKING_ENABLE_EXTERNAL_TOOLS or external feature flags.
• If you want to expose only the retrieval capabilities to clients, consider using connector mode (SMART_THINKING_MODE=connector) alongside a transport like HTTP/SSE.
• For deployment, a minimal footprint is achievable by combining --transport=http --mode=connector with a reverse proxy (ngrok, fly.io, render, etc.).
• The server exposes a public endpoint for registry discovery at /.well-known/mcp/server-card.json and supports a robust local logging/trace pipeline for auditing reasoning paths.
• Development workflows include npm run build, npm run lint, npm run test, and npm run test:coverage to keep the codebase healthy before releases.
• If external tools are enabled, remember to manage and protect any external service integrations and keys; by default, external LLMs and embeddings are disabled.